Update app.py
Browse files
app.py
CHANGED
@@ -7,34 +7,94 @@ from langchain.document_loaders import PyPDFLoader
|
|
7 |
from langchain.text_splitter import CharacterTextSplitter
|
8 |
|
9 |
# Load environment variables
|
10 |
-
load_dotenv()
|
11 |
|
12 |
-
# Load and process the PDF files
|
13 |
-
loader = PyPDFLoader("./new_papers/ALiBi.pdf")
|
14 |
-
documents = loader.load()
|
15 |
|
16 |
-
# Split the documents into chunks and embed them using HuggingFaceBgeEmbeddings
|
17 |
-
text_splitter = CharacterTextSplitter(chunk_size=100, chunk_overlap=0)
|
18 |
-
vdocuments = text_splitter.split_documents(documents)
|
19 |
|
20 |
-
|
21 |
-
|
|
|
22 |
|
23 |
-
|
24 |
-
|
25 |
-
|
26 |
-
|
27 |
-
|
28 |
-
|
29 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
30 |
|
31 |
-
|
32 |
-
api_db = FAISS.from_texts(texts=docs_text, embedding=embeddings)
|
33 |
|
34 |
# Define the PDF retrieval function
|
35 |
def pdf_retrieval(query):
|
36 |
# Run the query through the retriever
|
37 |
response = api_db.similarity_search(query)
|
|
|
38 |
return response
|
39 |
|
40 |
# Create Gradio interface for the API retriever
|
|
|
7 |
from langchain.text_splitter import CharacterTextSplitter
|
8 |
|
9 |
# Load environment variables
|
10 |
+
#load_dotenv()
|
11 |
|
|
|
|
|
|
|
12 |
|
|
|
|
|
|
|
13 |
|
14 |
+
def get_pdf_text(pdf_docs):
|
15 |
+
"""
|
16 |
+
Extract text from a list of PDF documents.
|
17 |
|
18 |
+
Parameters
|
19 |
+
----------
|
20 |
+
pdf_docs : list
|
21 |
+
List of PDF documents to extract text from.
|
22 |
+
|
23 |
+
Returns
|
24 |
+
-------
|
25 |
+
str
|
26 |
+
Extracted text from all the PDF documents.
|
27 |
+
|
28 |
+
"""
|
29 |
+
text = ""
|
30 |
+
for pdf in pdf_docs:
|
31 |
+
pdf_reader = PdfReader(pdf)
|
32 |
+
for page in pdf_reader.pages:
|
33 |
+
text += page.extract_text()
|
34 |
+
return text
|
35 |
+
|
36 |
+
|
37 |
+
def get_text_chunks(text):
|
38 |
+
"""
|
39 |
+
Split the input text into chunks.
|
40 |
+
|
41 |
+
Parameters
|
42 |
+
----------
|
43 |
+
text : str
|
44 |
+
The input text to be split.
|
45 |
+
|
46 |
+
Returns
|
47 |
+
-------
|
48 |
+
list
|
49 |
+
List of text chunks.
|
50 |
+
|
51 |
+
"""
|
52 |
+
text_splitter = CharacterTextSplitter(
|
53 |
+
separator="\n", chunk_size=1500, chunk_overlap=300, length_function=len
|
54 |
+
)
|
55 |
+
chunks = text_splitter.split_text(text)
|
56 |
+
return chunks
|
57 |
+
|
58 |
+
|
59 |
+
def get_vectorstore(text_chunks):
|
60 |
+
"""
|
61 |
+
Generate a vector store from a list of text chunks using HuggingFace BgeEmbeddings.
|
62 |
+
|
63 |
+
Parameters
|
64 |
+
----------
|
65 |
+
text_chunks : list
|
66 |
+
List of text chunks to be embedded.
|
67 |
+
|
68 |
+
Returns
|
69 |
+
-------
|
70 |
+
FAISS
|
71 |
+
A FAISS vector store containing the embeddings of the text chunks.
|
72 |
+
|
73 |
+
"""
|
74 |
+
model = "BAAI/bge-base-en-v1.5"
|
75 |
+
encode_kwargs = {
|
76 |
+
"normalize_embeddings": True
|
77 |
+
} # set True to compute cosine similarity
|
78 |
+
embeddings = HuggingFaceBgeEmbeddings(
|
79 |
+
model_name=model, encode_kwargs=encode_kwargs, model_kwargs={"device": "cpu"}
|
80 |
+
)
|
81 |
+
vectorstore = FAISS.from_texts(texts=text_chunks, embedding=embeddings)
|
82 |
+
print("-----")
|
83 |
+
print(vectorstore.as_retriever.similarity("What is ALiBi?"))
|
84 |
+
print("-----")
|
85 |
+
return vectorstore
|
86 |
+
|
87 |
+
pdf_text = get_pdf_text("./new_papers/ALiBi.pdf")
|
88 |
+
text_chunks = get_text_chunks(pdf_text)
|
89 |
+
api_db = get_vectorstore(text_chunks)
|
90 |
|
91 |
+
|
|
|
92 |
|
93 |
# Define the PDF retrieval function
|
94 |
def pdf_retrieval(query):
|
95 |
# Run the query through the retriever
|
96 |
response = api_db.similarity_search(query)
|
97 |
+
print(response)
|
98 |
return response
|
99 |
|
100 |
# Create Gradio interface for the API retriever
|